AI Coding Tools Setup Guide
This guide explains how to configure AI coding assistants (Claude Code and GitHub Copilot) to work effectively with the GWS Core development environment.
Overview
The GWS Core environment provides specialized commands and instructions that help AI coding assistants understand the unique architecture of this platform. These commands enable AI assistants to:
- Generate Constellab Tasks (data processing components)
- Create Streamlit web applications
- Build Reflex web applications
- Follow GWS Core best practices and conventions
What are AI coding tools?
AI coding tools like Claude Code and GitHub Copilot are AI assistants that help you write code faster. They can:
- Generate code based on your descriptions
- Explain existing code
- Debug issues
- Suggest improvements
- Follow project-specific guidelines
Custom Commands & Instructions
What are Custom Commands?
Custom commands are specialized instructions that teach AI assistants about your project's specific requirements, architecture, and best practices. They act as expert guides that help the AI:
- Understand your environment - GWS Core architecture, file structure, and conventions
- Generate correct code - Following established patterns and best practices
- Use the right tools - CLI commands, testing frameworks, and development workflows
- Avoid common mistakes - By encoding domain expertise into reusable prompts
How Commands Work
When you invoke a command (e.g., /gws-task-expert Create a table transformer), the AI assistant:
- Loads the specialized instruction file
- Receives context about GWS Core architecture
- Understands what you're trying to build
- Generates code following best practices
- Knows how to test and validate the result
Benefits of Using Custom Commands
- Faster development - No need to explain GWS Core concepts every time
- Consistent code quality - AI follows the same patterns across your codebase
- Reduced errors - Built-in knowledge of common pitfalls
- Better documentation - AI generates properly formatted docstrings and comments
- Easier onboarding - New developers get AI assistance that knows your project
Claude Code Setup
Claude Code is Anthropic's official AI coding assistant that runs in your terminal or editor.
Prerequisites
Installation
To install Claude Code and configure it with all available GWS commands: gws claude init
This command installs Claude Code and configures it with the available commands.
Using Claude Code
Open claude code in your terminal (claude) or claude code extension in your editor.
Use the / prefix to invoke commands: /gws-<command-name> [your detailed description]
Verifying Installation
Check installed commands: gws claude commands --list
Check Claude Code version: claude --version
GitHub Copilot Setup
GitHub Copilot is GitHub's AI coding assistant integrated into popular editors.
Prerequisites
- GitHub account with Copilot access
- GitHub Copilot extension installed in VSCode.
Installation
Pull GWS commands to the global GitHub Copilot prompts folder: gws copilot commands --pull
This command will copy GWS commands to /lab/user/.github/prompts/.
Using GitHub Copilot
In the Copilot chat window, use the / prefix to invoke commands: /gws-<command-name> [your detailed description]
⚠️ Note: ensure that VsCode is opened in folder /lab/user to allow Copilot to have access to commands.
Verifying Installation
List available GWS commands: gws copilot commands --list
Available Commands
1. Task Expert (/gws-task-expert)
Description: Create or modify a Constellab Task that processes data resources
Use Cases:
- Creating new data processing tasks
- Modifying existing tasks
- Adding inputs/outputs/configuration
- Writing task tests
Example:
/gws-task-expert Create a TableTransposer task that takes a Table as input and returns the transposed Table as output
What it provides:
- Task decorator setup
- Input/output specifications
- Configuration parameters
- Run method implementation
- Logging and progress tracking
- Unit tests with TaskRunner
- Documentation in markdown
2. Streamlit App Developer (/gws-streamlit-app-developer)
Description: Create, develop, modify, or debug a Streamlit web application
Use Cases:
- Building interactive data apps
- Creating dashboards
- Implementing data visualization
- Adding UI components
Example:
/gws-streamlit-app-developer Create an app that allows users to upload a CSV file, display it in a table, and generate basic statistics
What it provides:
- Complete app structure
- State management with session_state
- Custom GWS Core components
- Development configuration
- Testing instructions
- Best practices for caching and performance
3. Reflex App Developer (/gws-reflex-app-developer)
Description: Create, develop, modify, or debug a Reflex web application
Use Cases:
- Building full-stack Python web apps
- Creating multi-page applications
- Implementing complex state management
- Adding routing and navigation
Example:
/gws-reflex-app-developer Create a multi-page app with a home page, data upload page, and visualization page
What it provides:
- Reflex app structure
- State management classes
- Page routing
- Component composition
- Development workflow
- Testing and debugging guidance
Best Practices
When Using AI Coding Tools
- Be Specific in Your Requests❌ "Create a task"✅ "Create a task that filters a Table resource based on a column value threshold"
- Mention Inputs and OutputsSpecify resource types (Table, File, etc.)Describe expected output formatInclude any configuration parameters needed
- Reference Existing Patterns"Similar to TableTransposer in gws_core""Following the pattern in impl/table/tasks"
- Request TestsAsk for unit tests explicitly if not generatedSpecify test scenarios to cover
- Iterate and RefineReview generated codeAsk for modifications or improvementsRequest explanations for unclear parts
Keeping Commands Up to Date
Commands are automatically updated on lab startup. You can manually update them as needed.
# For Claude Code
gws claude update
# For GitHub Copilot
gws copilot update
